Decentralized Learning in Healthcare: A Review of Emerging Techniques

Recent developments in deep learning have contributed to numerous success stories healthcare. The performance of a model generally improves with the size training data. However, there are privacy, ownership, and regulatory issues that prevent combining medical data into traditional centralized storage. Decentralized approaches enable collaborative by distributing process among several nodes or devices. Conceptually, decentralized builds on earlier work distributed optimization, but focus this paper is recent emerging techniques such as Federated Learning (FL), Split (SL), hybrid Split-Federated (SFL). With common, universal models aggregator servers, FL overcomes difficulties training. Additionally, patient remains at local party, upholding security anonymity SL enables machine without directly accessing clients end It further enhances privacy setting mitigates clients’ storage issues. In survey, we first provide contemporary survey FL, SL, SFL approaches. Second, discuss their state-of-the-art applications healthcare, particularly image analysis. Third, review these under challenging conditions statistical system heterogeneity, preservation, communication efficiency, fairness, etc. Then, address existing tackle challenges. We detail unique complications related healthcare including data, security, Finally, outline potential areas for research developing personalized models, reducing bias, incorporating non-IID features, hyperparameter tuning, sufficient incentive mechanisms, domain expertise knowledge.